ThinkOmni: Lifting Textual Reasoning to Omni-modal Scenarios via Guidance Decoding

Omni-modal reasoning is essential for intelligent systems to understand and draw inferences from diverse data sources. While existing Omni-modal Large Language Models (OLLM) excel at perceiving diverse modalities, they lack the complex reasoning abilities of recent Large Reasoning Models (LRM). However, enhancing the reasoning ability of OLLMs through additional training presents significant challenges, including the need for high-quality data, task-specific adaptation, and substantial computational costs. To address these limitations, we propose ThinkOmni, a training-free framework that lifts textual reasoning to omni-modal scenarios. ThinkOmni introduces two key components: 1) LRM-as-a-Guide, which leverages off-the-shelf LRMs to guide the OLLM decoding process; 2) Stepwise Contrastive Scaling, which adaptively balances perception and reasoning signals without manual hyperparameter tuning. Experiments on six multimodality reasoning benchmarks demonstrate that ThinkOmni consistently delivers performance improvements, with main results achieving 70.2% on MathVista and 75.5% on MMAU. Overall, ThinkOmni offers a flexible and generalizable solution for omni-modal reasoning and provides new insights into the generalization and application of reasoning capabilities. Code is publicly available at https://github.com/1ranGuan/thinkomni Actually, this is not a trivial problem, and despite considerable efforts, existing approaches to omnimodal reasoning are still limited in several critical aspects. Specifically, 1) Insufficient modality diversity. Current studies largely focus on specific modalities (e.g., image (Liu et al., 2025b;a; Lin et al., 2025 ), audio (Li et al., 2025a), or video (Wang et al., 2025)), rather than generalizing across Work done at Xiaomi Inc.